![](https://steemitimages.com/DQmPxPsUXD9goSXNXBg7Fw6EM24MMRoJnMd8S2Re9GSZ4eZ/image.png)

Researchers at Stanford University's Slack particle accelerator have found a way to speed up the analysis of gravitational lens images, usually analyzed for weeks and months.

Gravitational lens A cosmic phenomenon predicted by the world Albert Einstein is the result of a dense gathering of stars working on the curvature of light from a smaller body beyond, such as a galaxy. The focus of light is in a focus that allows the light to be seen magnified, as if the light passes through an ordinary glass lens, the size of the light body increases and appears larger than it actually is.

This is a useful phenomenon as scientists have helped to discover the outer planets, understand the evolution of galaxies, monitor galaxies and discover black holes. But analyzing images that are affected by the gravity lens takes too long, and requires researchers to compare real images with computer simulation images. Analysis of the effect of a single attractive lens may take weeks or months.

Thanks to their discovery, the researchers reduced the analysis time of the gravitational lens images to parts of a second. The team trained a computer neural network using half a million images of an attractive lens.

"It's exciting that neural networks learn for themselves what characteristics you're looking for," said project researcher Phil Marshall. "It's like the way young children learn to recognize things. You do not tell them what the dog is, just show them pictures of dogs. "He said.

"It's not just as if they chose pictures of dogs from a photo collection, but also as if they provided information about dogs, such as their weight, length and age," said Yashar Hazafah, another project researcher.

With the construction of new space telescopes that can detect more examples of gravity lenses, such methods will be necessary to quickly search data. More importantly, neural network analyzes can be done using a laptop or even a mobile phone.

Source:
https://www.technologyreview.com/the-download/608772/ai-lets-astrophysicists-analyze-images-10-million-times-faster/
https://www.dailydot.com/debug/nasa-fdl-ai-2017/